Technical Skills. Languages: Python, Julia, Rust, SQL. Natural Language Processing, Hierarchical Machine Learning models, State Space Models, Time Series modeling, AWS, Excel.
- B.A in Applied Mathematics | Boston University (2023- )
- B.S in Data Sciences | Boston University (2023- )
Regime Detection in Commodity Futures: An HMM Framework
- Using Volatility and event-driven data, built a regime switching trading strategy using Hidden Markov Models (HMMs) for identifying short-term regime shifts for commodity markets.
- Validated model using purged K-fold cross validation (1.5M+ rows of data). Live data scraping and cleaning functionality, and cached data for fast computation.
- Built robust time-series sampling pipeline from scratch; volatility bars, Directional-Change, time-based resampling. Handles 2M+ rows of data, developed multithreading functionality to reduce computation by 50%.
- Implemented full trading pipeline hosted on AWS EC2; Databento for market data, brokerage RestAPI requests.
Natural Language Processing Trading Bot
- Obtained university funding for a sentiment analysis trading system, managing a team of four. Fine tuned Bidirectional-Encoder-Transformer (BERT) neural net from Google Deepmind to 93% validation accuracy using 35,000 point self curated dataset.
- Constructed recurrent neural network (RNN) to uncover temporal relationship between news/sentiment and next day stock movements. 6% backtested returns over two months across portfolio of 20 large-cap equities. .
Portfolio Optimization of BUFC's 1.2M Long Equity Fund
- Implemented a 3d correlation graph to analyze time series correlations of the fund's holdings using the spearman coefficient.
- Designed position sizing algorithm for new positions that leverages a four factor model, idiosyncratic expected returns.
- Proposed trading plan for favorable equity weightings based on industry exposure metrics and weighting scheme aligned to the Russell 2000 Index (Fund Benchmark).
Analysis of Car Crashes in the United States (DS110 Final project on Github)
- Cleaned, normalized, and analyzed correlations of over 7 million car crashes using seaborn heatmaps and pandas
- Utilized a neural network built using tensorflow to predict the likelihood of a car crash based on several conditions (i.e. existence of an intersection, weather condition, ect.)
- Resulting Neural Net had 87% validation accuracy in predicting instances of car crashes.
- Created an interactive application using the plotly dash python library to view the density of car crashes per county and state in the United States.
California road graph analysis using Rust (On github as DS210 final project)
- Implemented graph algorithms such as Dijkstra's algorithm and Breadth-First Search (BFS) to find shortest paths and compute graph metrics using real-world data from the California road network.
- Analyzed graph properties, calculating graph density and clustering coefficients to understand connectivity and local structure.
- Conducted subgraph analysis, creating subsets of the graph to manage computational complexity and derive additional insights, including average distance calculations.
- Utilized Rust programming language to compile and execute the code, ensuring error handling and user input validation for accurate graph analysis.